owlbite

12 ms·

61.
by owlbite 2y ago
Yeah, definitely a common phrase in the UK, basically the sense is that if you put a (firework) rocket up someone's behind and light it, they're going to move pretty damn fast to rectify the situation before the fuse runs out.
62.
by owlbite 2y ago
If I remember correctly, real libm implementations rarely use an analytically derived approximation, they just fit a polynomial rational approximation to minimize the error on the required range and precision.
63.
by owlbite 2y ago
I find the same problem with gmail on my iPhone. Every time I clock a link it asks me to use Chrome or the default browser. I set a default browser for a reason dammit, just use that one.
64.
by owlbite 2y ago
Algorithms for producing the same stream of random numbers in serial or parallel have been well known for many years. Perhaps the easiest in concept is "encrypt the sequence 1, 2, 3, 4, 5, ...".
65.
by owlbite 2y ago
That's what counter-based RNGs are for.
66.
by owlbite 2y ago
Something looks dubious with the benchmarking here to me. Top-tier numerical linear algebra libraries hold all hit the same number (give or take a few percent) for matrix multiply, because they're all achieving the same hardware peak p
67.
by owlbite 3y ago
I'm told cash is actually more troublesome for merchants than electronic means, especially as more transactions become cashless. They have to pay for keeping change, physical security, transport of takings, bank cash processing fees, e
68.
by owlbite 3y ago
I think the main issue is auto tools tries to support so many different shells/versions all with their own foibles, so the resulting cross-compatible code looks obfuscated to a modern user. Something built on python won't cover qu
69.
by owlbite 3y ago
And here I was thinking the Fields Medal was Math's Top Prize.
70.
by owlbite 3y ago
For things where there is a significant "cloud" type cost I am happy enough paying for that. But the software cost for the code in the car at the time of purchase should be covered by the cost of the car. It's not like there
71.
by owlbite 3y ago
The thing that's annoying me right now is how Ford wants to sell you back features like their advanced cruise control and navigation systems. If my employment contract didn't make things legally murky I'd totally be trying to
72.
by owlbite 3y ago
I suspect that if you have a sequence w_0, ... w_n with the property but all possible remaining words in your sequence following w_n are the wrong color, then you can use set w_0' = concat(w_0, ... w_n) and declare victory.
73.
by owlbite 3y ago
I'd add an addendum - look for stuff you don't need to do (e.g. at a low-level can you avoid that memcpy, zeroing things or multiple allocations when one would do, but even better at a high-level if you can avoid doing the whole t
74.
by owlbite 3y ago
Focussing on vector length is an error. You should care about throughput and number of different instructions you can retire. If I have a core with four 128-bit neon vector units, I have the same throughput as an x86 with two AVX2 units or
75.
by owlbite 3y ago
Code portability isn't performance portability, a fact that was driven home back in the bad old OpenCL era. Code is going to have to rewritten to be efficient on AMD architectures. At which point why tie yourself to the competitor&#x27
76.
by owlbite 3y ago
Also bitfields / any application using the hi bit as a flag for something or other is going to break now.
77.
by owlbite 3y ago
This basically reflects reasons I went from academia to industry. There is no real IC career path in academia, if you want to get paid more than peanuts you basically spend your entire life applying for grants and managing other people to d
78.
by owlbite 3y ago
Also ML gets a lot more funding (academic or commercial) than MILP.
79.
by owlbite 3y ago
FEAs probably involve sparse matrices, which have a lot more complexity than simple dense matrices. For example compute optimal reordering of a generic sparse matrix is iirc NP-complete.
80.
by owlbite 3y ago
I'd say that the algorithm matters more. As in, today's mathematicians know more about the subject. I suspect the actual implementation of said algorithms probably achieves a lower % of peak performance than the older ones (though
81.
by owlbite 3y ago
Also worth observing BLAS is an even weirder case as odds are the CPU architecture designed to ensure they could hit a high % of peak FMA throughput for that kernel, so there's a lot of SW/HW co-design going on there too.
82.
by owlbite 3y ago
Like much of the advice "it depends" and "measure it". Basically hyper-threading shares a whole load of resources between threads. If your code ends up contending on those resources you probably run slower. If your code
83.
by owlbite 3y ago
We'll have to see. I could definitely see someone spending a lot of time training for a specific algorithmic kernel and microarchitecture and beating the best human results (by a few percent). I'd be very surprised if that can be
84.
by owlbite 3y ago
Right, but we still tend to sidestep the compiler and/or spend hours of human time tuning the input to get the right output for core kernels.
85.
by owlbite 3y ago
These ML-compilers are being overhyped. It's all the same trade-off as a traditional compiler: you get a lot more throughput than hiring a specialist performance programmer, but the latter will typically outperform, possibly by orders
86.
by owlbite 3y ago
As with any super-hyped technology, people want to use the latest and greatest to solve it, even when traditional methods are cheaper, easier, more reliable and more accurate. See also: crypto.
87.
by owlbite 3y ago
I'd be somewhat concerned about an overzealous officer with a gun who "stops" me before I can pull out the paperwork.
88.
by owlbite 3y ago
As far as I can see Epic doesn't have a problem with Apple/Google/Steam's business model. They have a problem with not being Apple/Google/Steam themselves.
89.
by owlbite 3y ago
I don't know if I'd want Linear Algebra proofs - they tend to veer into results over vector fields rather than the more useful "solve Ax=b" and "here is how AI works" parts (i.e we should teach real-world compu
90.
by owlbite 3y ago
Very much this, and low risk bugs compound at scale. If you're in a very large FANNG type company, and say you have 1000 components that each ship 1 bug each day that has a 0.1% chance of breaking something critical, that translates to

More ›